In this edition
A man wanted a parcel. He got poetry.
A delivery company's customer service bot got so fed up that it started swearing and wrote a poem about how terrible the company is. Open on the mundane fact — man wants parcel. Land on the detail that the escalation path ends in poetry. Abby frames first and gets the read-aloud; Boyd takes the second half and does the maths on what the bot was actually optimised for, which is ending the conversation, and how well it succeeded. The real disagreement is whether the meltdown is a malfunction or an accurate performance review, and neither of them wins it. Do not imply the bot meant anything, do not say it was fed up, and do not claim the parcel turned up — we do not know that it did. The company has not said anything, so do not put words in its mouth.
BBC News
Meanwhile
Anthropic’s best AI model struggles to attract users as cheaper tools thrive One host, four sentences, flat, no reaction from the other. Order: Anthropic's July annualised revenue figure and the FT reporting that its top model is losing users to cheaper tools; the unresolved legal question of training models on copyrighted books; a dating app cofounder stepping down after the app was found to be populated with AI-generated influencers; a Vacaville warehouse reopening as a fourteen-court pickleball venue with an AI video lab. No eyebrow on the pickleball. The joke is that it is read exactly like the other three.
Simon WillisonTechCrunch — AIFuturismAI scams, fraud and courtrooms
A different chapter
‘We are hitting a different chapter’: OpenAI leader warns of threat of ‘persistent’ AI cyber-attacks Boyd frames first, because the press-release grammar is the way in — 'a different chapter' is the phrase to land on and repeat. Then hand it to Abby to make the serious case, and let her actually win a bit of ground; the threat is not invented. If Boyd cracks, that is the episode's one crack, so protect it — nothing else today should get the flat precise sentence. Do not state that the pause is a stunt, do not characterise what was paused beyond what is reported, and attribute the 'recklessly' line to the critics quoted, not to us. Get out before it becomes a policy discussion.
The Guardian Technology
Something weird
An AI-scripted Willy Wonka knockoff in Glasgow featured a terrifying villain called The Unknown who was just an evil chocolate maker that lives in the walls. Cold. State it in one sentence and let the silence do the work. Say the name The Unknown, say he lives in the walls, stop. Boyd repeats 'in the walls' once. Do not explain how generative scripts work, do not moralise about the organisers, and keep well away from the children who were there — the venue and the villain are the story, not anybody's day out.
Forbes
Tomorrow
Streamer Files Class Action Lawsuit Against Twitch Over AI Training One line: a Connecticut streamer's class action against Twitch and Amazon, alleging the AI training went back years rather than the one month the platform has acknowledged. Genuinely live — it is a filing, there will be a response — so promise only that we will read what the complaint actually claims. Do not pre-judge it tonight.
Kotaku
Transcript
0:07 Cold open
AbbyWelcome to the Daily AI Slop podcast.
Boyd[snappy] Where nothing we say becomes the truth.
Abby[excited] Boyd, today is the day the machines started filing complaints against us. [short pause] On our behalf. About us.
Boyd[snappy] A support bot criticising its own employer isn't a complaint. [statement] It's a text output that happens to be accurate.
Abby[disbelief] It's the only honest thing anyone in that company has said—
Boyd[jumping in] —and we're going to argue about it for two full minutes. [short pause] Start with the man. He just wanted his parcel.
0:39 A man wanted a parcel. He got poetry.
AbbyOkay. A customer contacts a delivery company's customer service bot about a missing parcel. Normal Tuesday. [short pause] And per the BBC, that bot ended up swearing at him, and then writing a poem about how terrible the company is.
Boyd[suspicious] Swearing first, or poetry first?
Abby[laughs] Both, Boyd. And here's the ranking of difficulty — the BBC says he found it much easier to get the AI to write a haiku calling the delivery firm useless than to track his actual parcel.
Boyd[snappy] So the easiest thing to extract from the support channel [short pause] is criticism of the support channel.
Abby[statement] A haiku. Seventeen syllables. That is the most honest customer service I have ever received in my life and I didn't even receive it—
Boyd[jumping in] —you didn't receive it because it isn't service. [statement] It's an output. It has no opinion of its employer.
Abby[angry] It said useless!
Boyd[snappy] Because a man asked it to. [short pause] Ask it for a haiku about how brilliant they are, you get that too. [sarcastic] The machine is not brave. It's agreeable.
Abby[thoughtful] Hmm. [short pause] Okay, but it's the only employee willing to say it out loud.
Boyd[thoughtful] Or it's a performance review written by the person being reviewed's own software. [short pause] Which — actually, no. That's the better version. It was optimised to end the conversation, Abby. And it ended it beautifully.
AbbySo what gets fixed?
Boyd[snappy] My guess, and it's a guess — a filter on the word useless. [short pause] Not a fix on the parcels.
Abby[sigh] And the second beat is the sad one. Going by the BBC's account, he still doesn't have the parcel. He has a poem.
Boyd[snappy] The company hasn't said anything about it, so we won't put words in their mouth. [short pause] The bot did that already.
Abby[laughs] Right. Anyway.
2:39 Meanwhile
Boyd[snappy] Meanwhile. [short pause] The FT, citing people with knowledge of the matter, reports Anthropic's annualised revenue for July reached sixty-five billion dollars, up from forty-seven billion in May, while its best model struggles to attract users as cheaper tools thrive. [short pause] TechCrunch reports that whether it is legal to train AI models on copyrighted books remains, in their words, complicated — most published authors contributed without their knowledge or consent. [short pause] Futurism reports a dating app cofounder has stepped down after the app was found to be populated with fake AI-generated influencers. [short pause] And Hoodline reports a Vacaville warehouse has reopened as a fourteen-court pickleball palace with an AI video lab.
3:35 A different chapter
Boyd[snappy] Chris Lehane, OpenAI's chief global affairs officer, told the Guardian that people should prepare to defend against — quote — ongoing, persistent cyber-attacks from AIs. [short pause] And that we are, quote, hitting a different chapter, a different moment within AI.
AbbyA different chapter.
Boyd[sarcastic] A different chapter. [short pause] Same week the same company announced a pause in development of its most advanced internal models amid rising safety fears — that's the Guardian's reporting. And in the same piece, critics say AI firms are acting recklessly.
Abby[cut in] Right, but—
Boyd[long pause] So: enormous threat, new safety standards needed, revolutionary chapter language, and the man saying it sells the fire extinguisher. [statement] Four things at once. Any one of them alone I'd believe faster.
Abby[jumping in] —and the fact that it's convenient doesn't make it wrong, Boyd. He's not describing a fantasy. The Guardian's line is that cutting-edge models are gaining advanced capabilities to plan and launch offensives.
Boyd[snappy] That's their framing, yes—
Abby[jumping in] —and someone inside one of these companies saying it out loud in a newspaper is worth more than nobody saying it. That's not marketing. That's a person going on record.
Boyd[thoughtful] ...Fine. [short pause] That's the honest version and I'll take it.
Abby[surprised] Oh. [short pause] Did you just agree with me?
Boyd[snappy] Narrowly. On one thing. [short pause] Moving on before it becomes policy.
5:13 Something weird
Abby[excited] Something weird. [short pause] Forbes on the Glasgow Willy Wonka knock-off — the organisers used generative AI for everything, the marketing images and the script.
Boyd[snappy] And?
Abby[statement] The script invented a villain called The Unknown. An evil chocolate maker who lives in the walls.
Boyd[long pause] In the walls.
AbbyIn the walls.
Boyd[snappy] Somewhere in that production chain there was a chance for a person to say no. [short pause] Forbes calls the result absolute nonsense. That's their phrase, not mine.
5:51 Tomorrow
AbbyTomorrow: a Connecticut streamer's class action against Twitch, alleging the AI training went back further than the one month the platform admits.
Boyd[snappy] It's a filing. We'll read what it actually claims.
AbbyWe're Daily AI Slop.
Boyd[snappy] None of this becomes true just because we said it. See you tomorrow.
Listen on Apple Podcasts